...
2. Description of the event
In this evaluation we will focus on the 3-day period (3-5 January) and two regions of 1x1 degree size: One in north-eastern Sweden between Vidsel and Alvsbyn (centre point 65.7N, 20.6E) and one over central Estonia (centre point 58.8N, 25.5E).
The plots below show analyses of T850 and z500 from 31 December to 6 January.
...
The plot below shows the 2-metre temperature anomaly from ERA5 for 3 to 5 January. Northern Sweden, Finland and Estonia saw cold anomalies larger than 15 degrees for this 3-day period.
The plot below show daily mean (based on 00 and 12UTC) temperature over northern Sweden (land points inside 60-65N, 10-20E) based on ERA5. Daily values (thin), 7-day running mean (thick) and daily climatology (dashed). The time series starts 3 August and ends 8 January. Since the beginning of October, almost all 7-day running mean have been colder than normal for this region.
...
3. Predictability
3.1 Data assimilation
...
The plots below show the observation statistics (first guess and analysis departures) from the surface DA for 2-metre temperature for Vidsel and Lulea-Kallax Airport.
3.2 HRES
The plots below show analysis of 3-day temperature 3-5 January and forecasts from 1 December January from IFS with 9 km resolution the the DestinE run with 4.4 km resolution. The 4.4 km forecast has a smaller area with below -30C (brown).
The plots below show 24-hour forecasts (IFS with 9 km and 4.4 km) from 3 January 00UTC and observations. For this time instance one can also see that the 4.4km forecast was fewer extremely cold areas.
The plots below show analysis of 3-day temperature 3-5 January and forecasts from 30 December from IFS (ENS control), and machine-learning forecasts from AIFS, PanguWeather and Graphcast. The box outlines a region between Vidsel and Alvsbyn. All machine-learning models are much warmer than the IFS control forecast.
3.3 ENS
The plots below show EFI for 3-day temperature 3-5 January. All forecast from 28 December and onwards captured the general structure of the cold event.
The plot below shows the forecast evolution for the 3-day (3-5 January) 2-metre temperature in the 1x1 degree box in northern Sweden between Vidsel and Alvsbyn. The evolution plot includes:
Analysis - green dot
Mean of observations - green hourglass
Medium-range ensemble blue box-and-whisker
ENS control - red dot
IFS DestinE 4.4km - purple dot
AIFS (1 degree) - pink
Graphcast - grey
PangyWeather - cyan
The ensemble gained the signal of the extreme from 25 December to 27 December. The machine-learning models (especially PanguWeather) struggled with the magnitude of the extreme in the medium-range. All forecasts were warmer than the mean of the observations (based on the stations in Alvsbyn and Vidsel).
The plot below shows the forecast evolution for the 3-day (3-5 January) 2-metre temperature in the 1x1 degree box in northern Sweden between Vidsel and Alvsbyn, going back 30 days before the event. The evolution plot includes:
Analysis - green dot
Mean of observations - green hourglass
Extended-range ensemble purple box-and-whisker
Model climate based on extended-range (pink) based reforecasts with 1800 fields. Climate maximum as triangle (to be added).
The plot below shows the same as above but for a 1x1 degree box in Estonia.
3.4 Monthly forecasts
The plots below show forecasts of weekly temperature anomalies for 1-7 January.
...
3.5 Comparison with other centres
4. Experience from general performance/other cases
...
5. Good and bad aspects of the forecasts for the event
...
- Early signal for cold over northern Sweden in the extended-range. Might connect to cold autumn and start of the winter
- The stronger signal appeared between 25 to 27 December
- Machine-learning models struggled with this case
- Destine 4.4km somewhat warmer than 9 km version of IFS